dm raid: enhance attempt_restore_of_faulty_devices() to support more devices
authorHeinz Mauelshagen <heinzm@redhat.com>
Tue, 9 Aug 2016 12:55:35 +0000 (14:55 +0200)
committerMike Snitzer <snitzer@redhat.com>
Tue, 16 Aug 2016 20:22:24 +0000 (16:22 -0400)
attempt_restore_of_faulty_devices() is limited to 64 when it should support
the new maximum of 253 when identifying any failed devices. It clears any
revivable devices via an MD personality hot remove and add cylce to allow
for their recovery.

Address by using existing functions to retrieve and update all failed
devices' bitfield members in the dm raid superblocks on all RAID devices
and check for any devices to clear in it.

Whilst on it, don't call attempt_restore_of_faulty_devices() for any MD
personality not providing disk hot add/remove methods (i.e. raid0 now),
because such personalities don't support reviving of failed disks.

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
drivers/md/dm-raid.c

index 81ec772..b1c2518 100644 (file)
@@ -3391,11 +3391,19 @@ static void raid_postsuspend(struct dm_target *ti)
 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
 {
        int i;
-       uint64_t failed_devices, cleared_failed_devices = 0;
+       uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
        unsigned long flags;
+       bool cleared = false;
        struct dm_raid_superblock *sb;
+       struct mddev *mddev = &rs->md;
        struct md_rdev *r;
 
+       /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
+       if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
+               return;
+
+       memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
+
        for (i = 0; i < rs->md.raid_disks; i++) {
                r = &rs->dev[i].rdev;
                if (test_bit(Faulty, &r->flags) && r->sb_page &&
@@ -3415,7 +3423,7 @@ static void attempt_restore_of_faulty_devices(struct raid_set *rs)
                         * ourselves.
                         */
                        if ((r->raid_disk >= 0) &&
-                           (r->mddev->pers->hot_remove_disk(r->mddev, r) != 0))
+                           (mddev->pers->hot_remove_disk(mddev, r) != 0))
                                /* Failed to revive this device, try next */
                                continue;
 
@@ -3425,22 +3433,30 @@ static void attempt_restore_of_faulty_devices(struct raid_set *rs)
                        clear_bit(Faulty, &r->flags);
                        clear_bit(WriteErrorSeen, &r->flags);
                        clear_bit(In_sync, &r->flags);
-                       if (r->mddev->pers->hot_add_disk(r->mddev, r)) {
+                       if (mddev->pers->hot_add_disk(mddev, r)) {
                                r->raid_disk = -1;
                                r->saved_raid_disk = -1;
                                r->flags = flags;
                        } else {
                                r->recovery_offset = 0;
-                               cleared_failed_devices |= 1 << i;
+                               set_bit(i, (void *) cleared_failed_devices);
+                               cleared = true;
                        }
                }
        }
-       if (cleared_failed_devices) {
+
+       /* If any failed devices could be cleared, update all sbs failed_devices bits */
+       if (cleared) {
+               uint64_t failed_devices[DISKS_ARRAY_ELEMS];
+
                rdev_for_each(r, &rs->md) {
                        sb = page_address(r->sb_page);
-                       failed_devices = le64_to_cpu(sb->failed_devices);
-                       failed_devices &= ~cleared_failed_devices;
-                       sb->failed_devices = cpu_to_le64(failed_devices);
+                       sb_retrieve_failed_devices(sb, failed_devices);
+
+                       for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
+                               failed_devices[i] &= ~cleared_failed_devices[i];
+
+                       sb_update_failed_devices(sb, failed_devices);
                }
        }
 }